The ALPS ProjectOpen Source Software for Quantum Lattice Models

Simon Trebst, ETH Zürich

The ALPS co!aboratio"

http://alps.comp-phys.org

The ALPS collaboration

University of Tokyo, Japan• Synge Todo

ETH Zürich, Switzerland• Fabien Alet• Prakash Dayal• Mathias Körner• Guido Schmid• Simon Trebst• Matthias Troyer• Philipp Werner• Stefan Wessel

Universität Marburg, Germany• Reinhard Noack• Salvatore ManmanaUniversität Würzburg, Germany• Martin Jöstingmeier TU Braunschweig, Germany

• Andreas Honecker

Universität Bonn, Germany• Axel Grzesik

Université de Toulouse, France• Andreas LäuchliRWTH Aachen, Germany• Ulrich Schollwöck• Ian McCulloch

TU Graz, Austria• Franz Michel• Hans-Gerd Evertz

Algorithms and Libraries for Physics Simulations

The ALPS project

• open source libraries and simulation codes for strongly correlated quantum systems

• Quantum Monte Carlo• exact diagonalization• DMRG

• Motivation

• established algorithms• increased demand for reliable simulations from

theorists and experimentalists

Strongly correlated systems

➡ compare microscopic models to experiments

0

2

4

6

8

10

12

0 200 400 600

experiment

simulationχ (1

0−5 c

m3 /m

ol C

u)

T (K)

SrCu2O3

J'/kB = 929 KJ/kB = 1904 K

simulation

Strongly correlated systems

• BEC of ultra-cold atoms in optical lattices

➡ quickly test theoretical ideas

0 5 10 15 20r / a

0.0

0.2

0.4

0.6

0.8

1.0

1.2

n(r)

0.0

0.1

0.2

0.3

!lo

ca

l (r)

talk H28.004 by Stefan Wessel

Simulation of quantum lattice models

• The status quo• individual codes • model-specific implementations• growing complexity of methods

• ALPS• community codes• generic implementations• common libraries simplify code development• common file formats

Modern technologies

• “Starting late” allows us to use• generic programming in C++ ➡ fast & flexible codes

• standard C++ libraries• MPI and OpenMP for parallelization• XML / XSLT for file processing

The ALPS sources

schedulerlattice model observablesdomain specific libraries

QMC ED DMRGapplications ...

numerics random ublas IETL

C / Fortran MPILAPACKBLAS

general C++ XML/XSLTgraph serialization

The ALPS lattice libraryA lattice

<LATTICEGRAPH name = "square lattice"> <FINITELATTICE> <LATTICE dimension="2"/> <EXTENT dimension="1" size="L"/> <EXTENT dimension="2" size="L"/> <BOUNDARY type="periodic"/> </FINITELATTICE> <UNITCELL> <VERTEX/> <EDGE type=“1”> <SOURCE vertex="1" offset="0 0"/> <TARGET vertex="1" offset="0 1"/> </EDGE> <EDGE type=“2”> <SOURCE vertex="1" offset="0 0"/> <TARGET vertex="1" offset="1 0"/> </EDGE> </UNITCELL> </LATTICEGRAPH>

A model

<BASIS> <SITEBASIS name="spin"> <PARAMETER name="S" default="1/2"/> <QUANTUMNUMBER name="Sz" min="-S" max="S"/> </SITEBASIS></BASIS>

<OPERATOR name="Splus" matrixelement="sqrt(S*(S+1)-Sz*(Sz+1))"> <CHANGE quantumnumber="Sz" change="1"/></OPERATOR><OPERATOR name="Sminus" matrixelement="sqrt(S*(S+1)-Sz*(Sz-1))"> <CHANGE quantumnumber="Sz" change="-1"/></OPERATOR><OPERATOR name="Sz" matrixelement="Sz"/>

<HAMILTONIAN name="spin"> <BASIS ref="spin"/> <SITETERM> -h*Sz </SITETERM> <BONDTERM source="i" target="j”> Jxy/2*(Splus(i)*Sminus(j)+Sminus(i)*Splus(j))+ Jz*Sz(i)*Sz(j) </BONDTERM></HAMILTONIAN>

€

HXXZ =Jxz2

(Si+S j

− + Si−S j

+)i, j∑ + Jz Si

zS jz

i, j∑ − h S1

z

i∑

The ALPS model library

Simulations with ALPSLattice<LATTICEGRAPH name = "square lattice"> <FINITELATTICE> <LATTICE dimension="2"/> <EXTENT dimension="1" size="L"/> <EXTENT dimension="2" size="L"/> <BOUNDARY type="periodic"/> </FINITELATTICE> <UNITCELL> ... </UNITCELL> </LATTICEGRAPH>

Model<BASIS> <SITEBASIS name="spin"> <PARAMETER name="S" default="1/2"/> <QUANTUMNUMBER name="Sz" min="-S" max="S"/> </SITEBASIS></BASIS>

<HAMILTONIAN name="spin"> <BASIS ref="spin"/> <SITETERM> -h*Sz </SITETERM> <BONDTERM source="i" target="j”> Jxy/2*(Splus(i)*Sminus(j)+Sminus(i)*Splus(j)) + Jz*Sz(i)*Sz(j) </BONDTERM></HAMILTONIAN>

ParametersLATTICE = “square lattice”L = 100

MODEL = “spin”Jxy = 1Jz = 1h = 0

{ T = 0.1 }{ T = 0.2 }{ T = 0.5 }{ T = 1.0 }

Results

quantum system

Quantum Monte Carlo Exact diagonalization DMRG

Open source developmentof ALPS

• Collaborative development

• Why open source codes?

• License discussion

• Funding situation

The History

the dark ag#

QMC, basic libraries‘94

loop code‘97

‘02

SSE‘01

‘03

ALPS‘04

‘94 DMRG

Current applications

• Quantum Monte Carlo• stochastic series expansions (SSE), F. Alet, M. Troyer

• loop code for spin systems, S. Todo

• continuous time worm code, S. Trebst, M. Troyer

• Exact diagonalization• full and sparse, A. Honecker, A. Läuchli, M. Troyer

• DMRG• single particle, S. Manmana, R. Noack, U. Schollwöck

• interacting particles, I. McCulloch

Collaborative development

• Some 15 developers, mostly PhDs.

• technical infrastructure

• source code control system (CVS)

• mailing lists

• web page

• semi-annual workshops

Austria, France, Germany, Japan, Netherlands and Switzerland

Developer workshops

• Developers meet at semi-annual workshops

• review of finished projects

• discussion of current developments

• road-map for new developments in the next 6 months

• Workshop are essential to coordinate and drive the development.

Open source codes

• Why open source?

• community profits from improvements by individual developers

e.g. adding disorder to models

• simplify code developmentbuild on lattice, model, ... libraries

• non-experts can use sophisticated algorithms

• reproducibility of published scientific results

Open source codes

• Why should a researcher publish his codes?

• start new collaborations / networks

• name recognition

• citations

• peer pressure to contribute

• The “cite me” - license

scientific returnopen sources

The “cite me” - license

• Applications codes

• free for non-commercial applications

• based on GNU public license

• citation requirements

• Modifications / improvements of codes

• should be integrated into ALPS

• not obligatory to publish

The “cite me” - license

• Applications codes

• free for non-commercial applications

• based on GNU public license

• citation requirements

• Library codes

• less restrictive

• partially available under a free license

User workshops• Oak Ridge ‘03

• part of workshop on Wang-Landau sampling

• Lugano ‘04

• general lectures on numerical methods

• hands-on tutorials of applications

• direct feedback from users

• workshops/tutorials set deadlines for developments

Become a user yourself!http://alps.comp-phys.org

Funding situation

• Software development classified as infrastructure project ➡ no funding by research grants

• combination with research projects

• no professional code developers

Funding situation• Funding sources for infrastructure

• computer centers, Swiss National Supercomputer Center ORNL

• companies?

• think big: group grantcan fund a software developer

• open source development helpful

Eidgenössische Technische Hochschule

Zürich

Ecole polytechnique fédérale de ZurichPolitecnico federale di Zurigo

Swiss Federal Institute of Technology Zurich

Calcolo Scientifico

Swiss Center for

Centro Svizzero di

Scientific Computing

CSCS

SCSC

Annual Report 2001/2002

ETH Zurich, Centro Svizzero di Calcolo Scientifico (CSCS)

Via Cantonale, Galleria 2, 6928 Manno, Switzerland

Experiences• Advantages of open source development ☺

• improved code design• more flexibility• code review by fellow developers

• new collaborations, joint research projects ☺• new contacts to experimental groups ☺• more work ☹

ALPS workshop ‘04 Lugano, September 26 - October 1

Classical Monte CarloSpin DynamicsQuantum Monte CarloExact diagonalizationDMRG

F. AletG. BrownH.-G. EvertzC. KollathA. LäuchliI. McCulloch

R. NoackT. SchulthessS. TodoS. TrebstShan-Ho Tsai

Invited speakers Hands-on tutorials